Research ArticleDiagnosis of Urinary Tract Infections by Urine Flow Cytometry:Adjusted Cut-Off Values in Different Clinical Presentations

Sabine K. Schuh,1 Ruth Seidenberg,1,2 Spyridon Arampatzis,3 Alexander B. Leichtle ,4

Wolf E. Hautz,1 Aristomenis K. Exadaktylos,1 Clyde B. Schechter,5 and Martin Müller 1

1Department of Emergency Medicine, Inselspital, Bern University Hospital, University of Bern, Bern, Switzerland2Department of Anesthesiology, Inselspital, Bern University Hospital, University of Bern, Bern, Switzerland3Department of Nephrology, Hypertension and Clinical Pharmacology, Inselspital, Bern University Hospital, University of Bern,3010 Bern, Switzerland4Department of Clinical Chemistry and Insel Data Science Center (IDSC), Inselspital, Bern University Hospital, University of Bern,Bern, Switzerland5Department of Family & Social Medicine & Department of Epidemiology and Population Health, Albert Einstein Collegeof Medicine, Bronx, New York, USA

Correspondence should be addressed to Martin Müller; martinmyller@gmail.com

Received 11 October 2018; Revised 3 January 2019; Accepted 6 February 2019; Published 3 March 2019

Academic Editor: Giuseppe Biondi-Zoccai

Copyright © 2019 Sabine K. Schuh et al. This is an open access article distributed under the Creative Commons Attribution License,which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Background. Bacterium and leucocyte counts in urine can be measured by urine flow cytometry (UFC). They are used to predictsignificant bacterial growth in urine culture and to diagnose infections of the urinary tract. However, little information isavailable on appropriate UFC cut-off values for bacterium and leucocyte counts in specific clinical presentations. Objective. Todevelop, validate, and evaluate adapted cut-off values that result in a high negative predictive value for significant bacterialgrowth in urine culture in common clinical presentation subgroups. Methods. This is a single center, retrospective, observationalstudy with data from patients of the emergency department of Bern University Hospital, Switzerland, with suspected infectionsof the urinary tract. The patients presented with different symptoms, and urine culture and urine flow cytometry wereperformed. For different clinical presentations, the patients were grouped by (i) age (>65 years), (ii) sex, (iii) clinical symptoms(e.g., fever or dysuria), and (iv) comorbidities such as diabetes and immunosuppression. For each group, cut-off values weredeveloped, validated, and analyzed using different strategies, i.e., linear discriminant analysis (LDA) and Youden’s index, andwere compared with known cut-offs and cut-offs optimized for sensitivity. Results. 613 patients were included in the study.Significant bacterial growth in urine culture depended on clinical presentation and ranged from 32.3% in male patients to 61.5%in patients with urinary frequency. In all clinical presentations, the predictive accuracy of UFC leucocyte and UFC bacteriumcounts was good for significant bacterial growth in urine culture (AUC ≥ 0 88). The adapted LDA95 equations did not exhibitconsistently high sensitivity. However, the in-house cut-offs (test positive if UFC leucocytes > 17/μL or UFC bacteria > 125/μL)were highly sensitive (>90%). In female, younger, and dysuric patients, even higher cut-offs for UFC leucocytes (169/μL, 169/μL,and 205/μL) exhibited high sensitivity. Specificity was insufficient (<0.9) for all tested cut-offs. Conclusions. For various clinicalpresentations, significant bacterial growth in urine culture can be excluded if flow cytometry measurements give a bacterialcount of ≤125/μL or a leucocyte count of ≤17/μL. In female patients, dysuric patients, and patients younger than ≤65 years, theleucocyte cut-off can be increased to 170/μL.

1. Introduction

The prevalence and severity of urinary tract infections(UTI) depends on demographic characteristics, the clinical

presentation, and the individual medical history [1, 2].For instance, UTI is the second most prevalent infectionseen in women aged 65 years or over [3], with a gradualincrease with age [4]. In susceptible subgroups, such as

HindawiDisease MarkersVolume 2019, Article ID 5853486, 10 pageshttps://doi.org/10.1155/2019/5853486

immunosuppressed, diabetic, and geriatric patients, there is agreater risk of death and life-threatening complications suchas urosepsis [3, 5, 6]. In diabetic patients, the mortality attrib-uted to an infection is higher than that in the populationwithout diabetes [7]. Thus, susceptible subgroups requirespecial attention with respect to UTI.

The gold standard for the diagnosis of a UTI is stillsignificant bacterial growth in urine culture combined withthe clinical presentation of the patient [8]. However, theresults of a urine culture can take up to several days—usuallyat least 24 hours—before they become available, and aprompt decision on treatment has to be taken within ashorter time frame. Thus, the decision for empirical antibi-otic treatment is often based on clinical symptoms and morerapidly available laboratory tests, such as urine dipstick,microscopic examination, and, nowadays, urine flow cytom-etry [9]. All of these are considered useful tools for predict-ing significant bacterial growth in urine culture and aretherefore used to diagnose a UTI.

Urine flow cytometry (UFC) has the advantage that it isstandardized, less expensive than urine culture and micro-scopic examination, and rapidly available—especially incomparison to urine culture [10].

To rule out negative urine culture (nonsignificant bacterialgrowth), most studies have focused on fixed cut-offs deter-mined by UFC for counts of leucocytes and bacteria in theurine (UFC leucocytes and UFC bacteria, respectively) thathelp to decide whether culturing is rational and indicatedfor treatment [11–13]. Several diagnostic algorithms suggestan “optimal” cut-off for predicting significant bacterialgrowth in urine culture [14]. Thus, the diagnosis of a UTIremains challenging, as the individual medical history andcomorbidities of the patient might influence the diagnosticvalidity of fixed UFC cut-offs for predicting significant bacte-rial growth in urine culture. Only a few studies have evaluatedthe use of adjusted cut-offs for bacterium and leucocytecounts determined by UFC in specific clinical presenta-tions—such as greater age, sex, and typical symptoms suchas dysuria, renal insufficiency, diabetes, and other immuno-suppression [1, 15–17].

The aim of this study is thus to develop, validate, and eval-uate adapted cut-off values that result in a high negative predic-tive value for significant bacterial growth in urine culture incommon and important clinical presentation subgroups—suchas women, dysuric patients, diabetics, immunosuppressedpatients, or patients younger than 65 years of age.

2. Methods

2.1. Study Design. This is a single-center, retrospective, obser-vational study with data from patients of the emergencydepartment (ED) of Bern University Hospital, Switzerland.More than 35,000 patients per year are treated at our ED.The patients’ data were anonymized. The study wasapproved by the ethical committee of the canton of Berne(KEK: 2016-01298) and performed according to Swiss law.Individual informed consent was waived, as the data hadbeen anonymized.

2.2. Study Population, Definition of Patient ClinicalPresentation, and Data Extraction.This is a secondary analysisof a published data set, in which we developed generic tools topredict significant bacterial growth in urine culture from UFCmeasurements [14]. In the study presented here, we focus onpatient characteristics: (i) age (>65 years and ≤65 years); (ii)gender; (iii) fever (>38.0°C on presentation in accordance witha previous UFC study [15]); (iv) urinary symptoms such asdysuria, urinary frequency, and UTI-specific abdominal pain,including suprapubic pain, lower abdominal pain, or flankpain; (v) renal insufficiency (classified by the medical history);(vi) diabetic patients; and (vii) immunosuppressed patients.Immunosuppression was defined as cancer under chemother-apy, transplantation, or a current hematological or rheumato-logical disease.

The study population was obtained through a keywordsearch for “urine culture” in the medical file. The inclusioncriteria were as follows: age> 16 years, presentation at theED from January 7th, 2016, to July 31st, 2016, and urineculture and urine flow cytometry obtained within 24 hoursof presentation. Thus, we included all patients for whomthe attending physician decided to carry out urine flowcytometry and urine culture, as they were assumed to havesuspected UTI. The flow chart and reasons for exclusionare shown in Figure 1. The medical history and clinical pre-sentation are routinely recorded by the attending physician.The records were stored in our emergency department’smedical database (E-Care, ED 2.1.3.0, Turnhout, Belgium).

2.3. Laboratory Analysis. Urine culture (gold standard) wasobtained from a clean midstream/catheter urine at the EDand sent to the laboratory within two hours. After 24 h and48 h of incubation, the results of the urine culture wererecorded by the Department of Clinical Microbiology. Urineflow cytometry was performed by the Center of LaboratoryMedicine at Bern University Hospital (the Inselspital), anISO 17025-accredited laboratory. The UFC was performedwith the UX-2000 flow cytometer (Sysmex Corporation,Kobe, Japan). The UFC is fully automated, and resultscan be viewed online by the attending physician 30minafter the start of the analysis. The procedures for urinecollection, storage, and analysis are described in a previouspublication [14].

2.4. Outcomes. Significant bacterial growth in the urineculture, defined as at least 104 colony forming units (CFU)per mL, was taken as the gold standard.

2.5. Statistics. Statistical analysis was performed with Stata®13.1 (StataCorp, College Station, Texas, USA).

To validate the obtained cut-offs, the data set was ran-domly divided into training and validation sets in a 1 : 1 ratio.

The different clinical presentations were separately ana-lyzed as follows: First, the area (AUC) under a receiveroperating curve (ROC) and its 95% confidence interval(CI) were calculated following logistic regression. This wasthen used to determine the discriminatory performance ofUFC bacterium and UFC leucocyte counts for predicting

2 Disease Markers

significant bacterial growth in urine culture in the specificclinical presentation.

Second, the following adjusted cut-offs for bacterium andleucocyte counts in UFC were established for the training set:

(i) Cut-offs with the highest Youden’s index(sensitivity + specificity − 1)

(ii) Sensitivity-optimized LDA95 equations (test positive⇔ ln UFC leucocytes + 1 > a × ln UFCbacteria +1 – b); the parameters a and b are determinedthrough discriminant analysis and the optimizationcriterion that sensitivity should be above 0.95 inthe training set (for easier reading and to derive aformula valid for all UFC values, the formulapresented by Monsen and Ryden was slightlychanged [18])

(iii) Optimized sensitivity cut-offs defined as the highestbacterium or leucocyte count with a sensitivity ≥0 95; a test was defined as negative if both cut-offsare below the identified thresholds.

Third, these cut-offs were used in the validationsample to assess the diagnostic performance with respectto sensitivity, specificity, and positive and negative likeli-hood ratio (LR). The in-house cut-off for a positive testused at our hospital is UFC bacteria > 125/μL or UFCleucocytes > 17/μL; this was also validated. Last, the diag-nostic performance of the different cut-offs in the clinicalpresentations was compared among the different methods.

3. Results

In total, 613 (100%) patients with a median age of 60years (interquartile range: 46-75) were included in theanalysis (Figure 1).

Table 1 shows that more than half of the study popula-tion was aged <65 years old (55.6%). Immunosuppression

was found in 35.1%. Fever was present in 31.4%, while dia-betes was coded in 22.2% and renal insufficiency in 33.9%of the patients. Typical symptoms for a UTI were less common,with 18.3% dysuria, 12.7% urinary frequency, and 19.4% withUTI-specific abdominal pain (including lower abdominal pain,suprapubic pain, and flank pain). Three hundred seven (307)consultations were randomized to the training set and 306patients to the validation set. Significant bacterial growth inurine culture, defined as at least 104CFU/mL, was found inapproximately four out of ten patients (40.2%). E. coli wasidentified in 48.6% of those patients.

The most common urological diagnosis was possibleurogenital infection (29.5%). Other important dischargediagnoses were uncomplicated UTI (13.4%) and urosepsis(9.6%), as well as “no specific urological diagnosis” (25.6%,e.g., other infectious disease and respiratory problem).

3.1. Incidence of Significant Bacterial Growth in Urine Cultureand Predictive Accuracy of UFC Bacterium and UFCLeucocyte Counts. The incidence of significant bacterialgrowth in urine culture was higher in females and olderpeople (49.2% and 46.3%, respectively) than in males andyounger people (32.3% and 35.8%, respectively). Dependingon the clinical symptoms and comorbidities, the incidenceranged from 32.3% in male patients to 61.5% in patients withurinary frequency (Table 2). The incidence of significantbacterial growth in the urine culture in the different clinicalpresentations did not differ significantly between the trainingand validation sets (all p > 0 05).

The discriminatory accuracy values of UFC leucocyte andUFC bacterium counts to predict significant bacterial growthin urine culture measured by the AUC were at least 0.88 in allstudied clinical presentations. In female patients, the AUCwas 0.88 (95% CI: 0.85, 0.92) and in male patients 0.95(95% CI: 0.92, 0.97). In the other clinical presentations, the95% CI of the AUC overlapped, although the actual AUCvalues were slightly different (Table 2).

Exclusion (n = 280):(i) Urine flow cytometry missing (n = 121)

(ii) Urine culture missing (n = 159)

Patients identified through the keyword search “urine culture” with differentsemantic combinations/abbreviations in all medical reports from

7th January 2016–31st July 2016(n = 893)

Patients included in the analysis(n = 613)

Randomized split into training set (50%) and validation set (50%):(i) Training set (n = 307)

(ii) Validation set (n = 306)

Figure 1: Flowchart.

3Disease Markers

3.2. Adapted Cut-Offs with the Highest Youden’s Index. Thetraining sample was used to find adapted cut-offs forUFC bacterium and UFC leucocyte counts that maximizedYouden’s index in each clinical subgroup. Striking differ-ences in the cut-offs were found (Table 3). For example,if Youden’s cut-off was 74/μL for UFC leucocyte countsand 63/μL for UFC bacterium counts, Youden’s index inpatients with renal insufficiency was 0.5. In patients with-out diagnosed renal insufficiency, the Youden-based cut-offfor UFC leucocyte counts was 108/μL and for UFC bacteriumcounts 470/μL, giving Youden’s index of 0.7.

Maximal values of Youden’s index were associated withhigh-sensitivity values, which were all above 88.2%, withlower sensitivity (<90%) in diabetic and female patients inthe validation set. Specificity was lower, with a maximumof 88.3% (range: 46.2% in patients with UTI-specificabdominal complaints up to 88.3% in febrile patients).The high-sensitivity values were reflected by a small nega-tive likelihood ratio (<0.1), indicating a large decrease inthe posttest probability of significant bacterial growth, whilethe positive likelihood ratio was <5 in all clinical presenta-tions (slight increase in posttest probability) [19].

3.3. Adapted Cut-Offs with Optimized Sensitivity > 95%Compared with the In-House Cut-Offs. Two differentmethods were used to adapt cut-offs to high sensitivity. Thefirst approach was to use the adapted LDA95 equation. Thesecond approach was to optimize sensitivity (>95%) for eachgroup of patients—firstly for the UFC bacterium count andsecondly for the UFC leucocyte count—and then to combinethese cut-offs in a single test. A test was defined as positive ifeither the UFC leucocyte or the UFC bacterium cut-off wasexceeded. These two approaches gave different results for

diagnostic performance (Tables 4 and 5). When specificitywas moderate, the adapted LDA95 equations failed to showgood sensitivity (>90%) for several groups of patients,including male, older, and diabetic patients (Table 4).

For all clinical presentations, the in-house cut-offs (testpositive if UFC leucocytes > 17/μL or UFC bacteria > 125/μL) exhibited high sensitivity and good negative likelihoodratios (<0.1). In total, 184 of the 613 patients (30.0%) hadUFC leucocyte < 17/μL and UFC bacteria < 125/μL in theflow cytometry results. The adapted sensitivity cut-offvalues were similar to the in-house values. In female, youn-ger, and dysuric patients, high sensitivity was obtained evenwith higher cut-offs for UFC leucocyte counts (169/μL,169/μL, and 205/μL) and a similar value for UFC bacteriumcounts (103/μL). Specificity was low for all testedcut-offs—especially the in-house cut-off—and ranged from22.2% in patients with urinary frequency to 60.6% in malepatients (Table 5).

4. Discussion

This is the first study to focus on the development, evalua-tion, and validation of cut-off parameters for UFC bacteriumand UFC leucocyte counts in different clinical scenarios inan emergency department population. This can then be usedto predict significant bacterial growth in urine culture, whichin turn indicates a UTI and thus the necessity of antibiotictreatment. The study covered patients of different sex, age,symptoms (UTI-specific abdominal pain, fever, urinary fre-quency, and dysuria), and comorbidities (diabetes, renalinsufficiency, and immunosuppression).

In all studied clinical presentations, the discriminativeaccuracy of UFC leucocyte and UFC bacterium counts for

Table 1: Patient characteristics (n = 613).

Characteristics (n (%)) Total (n = 613) Training set(n = 307)

Validation set(n = 306)

Sex

Male 316 (51.5) 157 (51.1) 159 (52.0)

Female 297 (48.5) 150 (48.9) 147 (48.0)

Age

>65 years 272 (44.4) 130 (42.4) 142 (46.4)

≤65 years 341 (55.6) 177 (57.6) 164 (53.6)

Clinical presentation

Dysuria 112 (18.3) 56 (18.2) 56 (18.3)

Urinary frequency 78 (12.7) 40 (13.0) 38 (12.4)

Fever (>38.0°C)1 192 (31.4) 96 (31.4) 96 (31.5)

Urinary tract infection-specific abdominal pain2 119 (19.4) 61 (19.9) 58 (19.0)

Comorbidities

Diabetes mellitus 136 (22.2) 62 (20.2) 74 (24.2)

Renal insufficiency 208 (33.9) 112 (36.5) 96 (31.4)

Immunosuppression 215 (35.1) 105 (34.2) 110 (36.0)

Hospitalisation 451 (73.6) 225 (73.3) 226 (73.9)1Two values (one from the training and one from the validation set) were missing; thus, n = 611. 2This includes lower abdominal, suprapubic, and flank pain.

4 Disease Markers

significant bacterial growth in urine culture was good(AUC ≥ 0 88). With the in-house cut-offs, a test was definedas positive if UFC leucocyte counts were >17/μL or UFCbacterium counts > 125/μL and with this exhibited highsensitivity in all groups of patients. Compared with thecut-offs for optimized sensitivity of >95% in female, youn-ger, and dysuric patients, high sensitivity and even higherspecificity were obtained with higher cut-offs for UFCleucocyte counts (169/μL, 169/μL, and 205/μL).

Although this is sometimes neglected in clinical prac-tice, the posttest probability of a disease depends on theincidence of a disease in the specific clinical setting andthe likelihood ratio (LR) of the test [20]. Given a specificclinical presentation with a specific symptom A, the pretestprobability, the incidence of the disease, can be convertedto the odds for the disease. The odds multiplied by theLRA(+) of the disease in the specific presentation withsymptom A and the general LRTest(+) leads to the posttest

odds, which can be transformed to the posttest probability[19]. Posttest probability is of great help in the in-depthinterpretation of test results [21]. However, this constructis not of great use in clinical practice, as it requires thespecific LR(+) for each symptom as well as for symptomcombinations and demands complex calculations [19].Nevertheless, it is important that clinicians understand theprimary concept of the pre- and the posttest probability.An alternative and more useful concept is the use ofadapted cut-offs for diagnostic tests, which are, for instance,used in the diagnosis of venous embolism with age-adjustedD-dimers [22]. To rule out a diagnosis, it is necessary thatthe adapted cut-offs consider the possible higher or lowerincidence of the disease in specific subgroups; otherwise,the test may not be valid in the specific presentation.

In our study, the incidence of significant bacterial growthin urine culture ranged from 32.8% in febrile patients to61.5% in patients with urinary frequency (Table 2). These

Table 2: Discriminatory accuracy of UFC leucocyte and UFC bacterium counts measured by the area under the receiver operating curve ofthe logistic regression to predict significant bacterial growth in urine culture.

Total (n = 613) SBU (%) AUC (95% CI) Training set (n = 307) SBU (%) Validation set (n = 306) SBU (%)

Sex

Male 32.3 0.95 (0.92, 0.97) 29.9 34.6

Female 49.2 0.88 (0.85, 0.92) 48.7 49.7

Age

>65 years 46.3 0.95 (0.92, 0.97) 45.4 47.2

≤65 years 35.8 0.90 (0.86, 0.93) 34.5 37.2

Dysuria

Yes 54.5 0.88 (0.82, 0.94) 57.1 51.8

No 37.3 0.92 (0.90, 0.95) 35.1 39.6

Urinary frequency

Yes 61.5 0.88 (0.80, 0.96) 70.0 52.6

No 37.4 0.92 (0.90, 0.94) 34.5 40.3

Fever (>38.0°C)1

Yes 32.8 0.94 (0.90, 0.98) 28.1 37.5

No 43.9 0.91 (0.88, 0.94) 44.3 43.5

UTI-specific abdominalpain2

Yes 54.6 0.88 (0.81, 0.93) 54.1 55.2

No 37.0 0.92 (0.90, 0.95) 35.5 38.7

Diabetes

Yes 44.1 0.94 (0.90, 0.98) 41.9 46.0

No 39.4 0.91 (0.89, 0.94) 38.4 40.5

Renal insufficiency

Yes 39.9 0.93 (0.89, 0.96) 37.5 42.7

No 40.7 0.92 (0.89, 0.94) 40.0 41.4

Immunosuppression

Yes 33.5 0.92 (0.88, 0.96) 33.3 33.6

No 44.2 0.91 (0.89, 0.94) 42.1 46.4

Total 40.2 — 39.1 41.81Two values (one of training and one of the validation set) were missing; thus, n = 611. 2This includes lower abdominal, suprapubic, and flank pain.Abbreviations: AUC: area under the receiver operating curve; CI: confidence interval; SBU: significant bacterial growth in urine culture; UFC: urine flowcytometry; UTI: urinary tract infection.

5Disease Markers

figures are comparable to the incidence of significant bacte-rial growth in urine culture in other studies [23]. The rela-tively low incidence of significant bacterial growth in urinecultures in immunosuppressed and febrile patients may beexplained by the relatively broad indication for obtaining aurine culture in these patients, even when there is no greatsuspicion of a urological infection.

As suggested by different studies [20, 24, 25], one strategyfor establishing cut-off values is to maximize Youden’s index.Our results suggest that this strategy does not lead to usefulcut-offs in many clinical presentations, as the resultingcut-offs can neither be used to “rule in” (high LR(+)) norbe used to “rule out” (LR(-)) significant bacterial growth inthe urine culture but lead to balanced cut-offs with regardto LR(+)/LR(-). In general, cut-offs established by maximiz-ing Youden’s index should be treated cautiously, and thedetailed diagnostic performance of such tests should be dem-onstrated. More adequate approaches to establish “optimal”

cut-offs require estimating the “costs” of misdiagnoses, falsepositives, and false negatives [26], a task which is hard toaccomplish. Thus, we focused on ruling out the diagnosis inthe further analyses by maximizing sensitivity. The suggestedlinear discriminant analysis, LDA95 [18], is not very feasiblein everyday use, as a calculator or computer is needed to eval-uate the results of a urine culture and the results found in thisstudy were not convincing. We slightly modified the formulapresented by Monsen and Ryden [18], in order to assure thatthe formula was valid for all numbers of urine flow bacteriumand leucocyte counts; this had no impact on the outcome.Adapted cut-offs for the clinical presentations were thereforedeveloped by maximizing the sensitivity of both UFC leuco-cyte and bacterium counts and combining these into one test.In the setting of immunosuppression, for instance, lowercut-offs for UFC bacterium and UFC leucocyte counts werefound to lead to high sensitivity (>95%). Stefanovic et al. alsostudied the diagnostic performance of urine flow cytometry

Table 3: Established cut-off values for UFC bacterium and leucocyte counts per μL with the highest Youden’s index and the correspondingdiagnostic performance in the validation sample in predicting significant bacterial growth in urine culture in different clinical scenarios.

Subgroup (no. in validation set)Highest Youden’s

indexDiagnostic performance using the established cut-offs∗

Lc (/μL) Bact (/μL) Sensitivity (%) (95% CI) Specificity (%) (95% CI) LR(+) LR(-)

Sex

Male (n = 159) 74 71 94.5 (84.9, 98.9) 71.2 (61.4, 79.6) 3.28 0.08

Female (n = 147) 169 2883 89 (79.5, 95.1) 79.7 (68.8, 88.2) 4.39 0.14

Age

>65 (n = 142) 57 63 95.5 (87.5, 99.1) 60.0 (48.0, 71.1) 2.39 0.07

≤65 (n = 164) 169 150 96.7 (88.7, 99.6) 64.1 (54.0, 73.3) 2.69 0.05

Dysuria

Yes (n = 56) 205 87 96.6 (82.2, 99.9) 66.7 (46.0, 83.5) 2.90 0.05

No (n = 250) 74 441 93.3 (87.3, 97.7) 74.8 (67.1, 81.5) 3.73 0.08

Urinary frequency

Yes (n = 38) 108 409 100 (83.2, 100) 55.6 (30.8, 78.5) 2.25 —

No (n = 268) 74 83 96.3 (90.8, 99) 60.6 (52.6, 68.2) 2.45 0.06

Fever (>38.0°C)1

Yes (n = 96) 108 527 91.7 (77.5, 98.2) 88.3 (77.4, 95.2) 7.86 0.09

No (n = 209) 171 99 94.5 (87.6, 98.2) 61.0 (51.6, 69.9) 2.42 0.09

UTI-specific abdominal pain2

Yes (n = 58) 169 128 96.9 (83.8, 99.9) 46.2 (26.6, 66.6) 1.80 0.07

No (n = 248) 69 83 95.8 (89.7, 98.9) 59.9 (51.6, 67.7) 2.40 0.07

Diabetes

Yes (n = 74) 76 1061 88.2 (72.5, 96.7) 77.5 (61.5, 89.2) 3.92 0.15

No (n = 232) 137 441 96.8 (91.0, 99.3) 77.5 (69.7, 84.2) 4.31 0.04

Renal insufficiency

Yes (n = 96) 74 63 95.1 (83.5, 99.4) 56.4 (42.3, 69.7) 2.18 0.08

No (n = 210) 108 470 94.3 (87.1, 98.1) 76.4 (67.9, 83.6) 4.00 0.08

Immunosuppression

Yes (n = 110) 74 127 97.3 (85.8, 99.9) 72.6 (60.9, 82.4) 3.55 0.04

No (n = 196) 151 470 94.5 (87.6, 98.2) 74.3 (64.8, 82.3) 3.68 0.071One patient in the validation set did not have sufficient data to determine if temperature > 38 0°C. 2This includes lower abdominal, suprapubic, and flank pain.Abbreviations: AUC: area under the receiver operating curve; Bact: UFC bacterium counts; CI: confidence interval; Lc: UFC leucocyte counts; LR: likelihoodratio; SBU: significant bacterial growth in urine culture; UFC: urine flow cytometry; UTI: urinary tract infection.

6 Disease Markers

in immunosuppressed patients. For high sensitivity, theyemployed even lower cut-offs for UFC bacterium counts ≥20/μLor UFC leucocyte counts > 5/μL [17]. Even with thislow cut-off, the sensitivity was lower (90.6%) for immuno-suppressed patients than for nonimmunosuppressedpatients. This is in contrast to our study, in which the highercut-off was sufficient to rule out significant bacterial growthin urine culture. In another study, the use of age-specificUFC parameters was recommended [16]; it was found thatfor people aged ≥65 years, a single cut-off of UFC bacteriumcounts of 200/μL leads to very high sensitivity and high spec-ificity (99.1% and 91.6%, respectively) with a minimal rate of0.87% false negative results.

Manoni et al. suggested using UFC leucocytes > 17/μLor UFC bacteria > 125/μL to define a pathological test[12]. Our study shows that these cut-offs (Table 5, thein-house cut-offs) are valid for ruling out significant bacte-rial growth in urine culture in all studied clinical presenta-tions; this cut-off is therefore recommended as a safe

rule-out cut-off in the ED. With this high-sensitivity cut-off, as many as 30% of the urine cultures could have beenavoided in our study. Martín-Gutiérrez et al. showed thatan even higher cut-off of 200/μL for the UFC bacteriumcount could help to avoid the culture of up to 60.2% ofsamples in the elderly population (>65 years) [16].

However, with these cut-offs, specificity was lacking, andin some groups, such as females, people aged ≤65 years, anddysuric patients, even higher cut-offs for UFC leucocytecounts (169/μL, 169/μL, and 205/μL, respectively) can assurehigh sensitivity.

4.1. Strengths and Limitations. Our results were obtainedwith the UX-2000. However, as the UX-2000 series has thesame technology as the frequently used UF-1000i series tomeasure urine leucocyte and bacterium counts; our cut-offscan also be applied to the latter machines.

In contrast to many other studies that focused on UFCcut-off values, a particular strength of this study lies in the

Table 4: Established high-sensitivity test with the use of an adapted 95 percent linear discrimination analysis equation (LDA95) and itsvalidation. LDA95 equation: ln UFC leucocyte + 1 > A × ln UFCbacteria + 1 + B.

SubgroupLDA95-equat.coefficients

Diagnostic performance using the established cut-offs∗

A B Sensitivity (%) (95% CI) Specificity (%) (95% CI) LR(+) LR(-)

Sex

Male (n = 159) -4.6 21.2 89.1 (77.8, 95.9) 76.0 (66.6, 83.8) 3.71 0.14

Female (n = 147) -3.4 18.9 95.9 (88.5, 99.1) 51.4 (39.4, 63.1) 1.97 0.08

Age

>65 (n = 142) -20.4 86.8 89.6 (79.7, 95.7) 70.7 (59.0, 80.6) 3.05 0.15

≤65 (n = 164) -1.6 10.5 95.1 (86.3, 99.0) 63.1 (53.0, 72.4) 2.58 0.08

Dysuria

Yes (n = 56) -1.4 10.4 96.6 (82.2, 99.9) 66.7 (46.0, 83.5) 2.90 0.05

No (n = 250) -4.5 20.8 91.9 (84.7, 96.4) 58.3 (50.0, 66.2) 2.20 0.14

Urinary frequency

Yes (n = 38) -2.3 17.8 100 (83.2, 100) 66.7 (41.0, 86.7) 3.00 —

No (n = 268) -4.2 19.7 90.7 (83.6, 95.5) 62.5 (54.5, 70.0) 2.42 0.15

Fever (>38.0°C)1

Yes (n = 96) -1.3 8.2 94.4 (81.3, 99.3) 68.3 (55.0, 79.7) 2.98 0.08

No (n = 209) -4.9 25.0 92.3 (84.8, 96.9) 66.1 (56.8, 74.6) 2.72 0.12

UTI-specific abdominal pain2

Yes (n = 58) -1.8 14.0 96.9 (83.8, 99.9) 57.7 (36.9, 76.6) 2.29 0.05

No (n = 248) -4.5 20.5 91.7 (84.2, 96.3) 63.8 (55.6, 71.4) 2.53 0.10

Diabetes

Yes (n = 74) -4.2 25.0 79.4 (62.1, 91.3) 80.0 (64.4, 90.9) 3.97 0.26

No (n = 232) -3.0 15.3 94.7 (88.0, 98.3) 61.6 (52.9, 69.7) 2.47 0.09

Renal insufficiency

Yes (n = 96) -2.4 13.0 87.8 (73.8, 95.9) 67.3 (54.3, 95.9) 2.68 0.18

No (n = 210) -3.9 21.0 93.1 (85.6, 97.4) 69.9 (61.0, 77.9) 3.10 0.10

Immunosuppression

Yes (n = 110) -1.0 6.5 97.3 (85.8, 99.9) 64.4 (52.3, 75.3) 2.73 0.04

No (n = 196) -5.9 31.2 91.2 (83.4, 96.1) 67.6 (57.8, 76.4) 2.82 0.131One patient in the validation set did not have sufficient data to determine if temperature > 38 0°C. 2This includes lower abdominal, suprapubic, and flank pain.Abbreviations: CI: confidence interval; LDA: linear discrimination analysis; LR: likelihood ratio; UTI: urinary tract infection.

7Disease Markers

external validity of our cohort. In their systematic review,Shang et al. concluded that most studies were laboratorybased. Thus, it was unclear whether they were generalizableto clinically suspected UTI [13]. On the other hand, our datawere collected in an ED population, and the results may notbe transferred to a non-ED patient population, such as ahealth care center.

Furthermore, trials are recommended to evaluate the useof clinically adapted cut-offs to determine the economicbenefit, as well as to quantify the impact of adapted cut-offson the prescription of (unnecessary) antibiotics.

The greatest limitation of our study is the retrospectivedesign. While the quality of the extracted data can beassured for both the exposure (UFC data) and the outcomedata (significant bacterial growth in urine culture), as theyare hard outcomes measured in certified laboratories, the

evaluated clinical presentations were predominantly definedthrough the electronic medical record. Thus, informationand misclassification bias through incomplete documenta-tion is an important potential limitation of this study. Thenumber of patients in some subgroups may have been toosmall to detect differences in the predictive accuracy in somespecific presentations (potentially underpowered statistics);this is reflected in the wide confidence intervals. Further-more, one particularly important subgroup of patients couldnot be studied with our retrospective study sample, i.e.,patients with a high suspicion of a UTI, namely, patientswith more than one symptom, for example, dysuria, fever,and suprapubic pain. As the sample size of this study wassmall, we suggest that our results should be verified in larger,prospective studies. As a consequence of the small samplesize, one cannot combine the results and apply them to

Table 5: Diagnostic performance of the in-house cut-offs and the sensitivity-optimized cut-offs for each clinical scenario.

Subgroup

Diagnostic performance using in-house cut-off(UFC leucocyte > 17/μL, UFC bacteria > 125/μL) Optimized cut-off corresponding diagnostic performance∗

Sensitivity (%)(95% CI)

Specificity (%)(95% CI)

LR(+) LR(-)Cut-offs(/μL)

Sensitivity (%)(95% CI)

Specificity (%)(95% CI)

LR(+) LR(-)Lc Bact

Sex

Male (n = 159) 96.4 (87.5, 99.6) 60.6 (50.5, 70.0) 2.44 0.06 74 17 100 (93.5, 100) 54.8 (44.7, 64.6) 2.21 —

Female (n = 147) 98.6 (92.6, 100) 37.8 (26.8, 49.9) 1.59 0.04 169 103 97.3 (90.5, 99.7) 45.9 (34.3, 57.9) 1.8 0.06

Age

>65 (n = 142) 95.5 (87.5, 99.1) 58.7 (46.7, 69.9) 2.31 0.07 57 21 100 (94.6, 100) 50.7 (38.9, 62.4) 2.03 —

≤65 (n = 164) 100 (94.1, 100.0) 45.6 (35.8, 55.7) 1.84 — 169 103 96.7 (88.7, 99.6) 60.2 (50.1, 69.7) 2.43 0.05

Dysuria

Yes (n = 56) 100 (88.1, 100) 33.3 (16.5, 54.0) 1.5 — 205 103 96.6 (82.2, 99.9) 66.7 (46.0, 83.5) 2.90 0.05

No (n = 250) 97 (91.4, 99.4) 54.3 (46, 62.4) 2.12 0.06 74 21 100 (96.3, 100) 45.7 (37.6, 54) 1.84 —

Urinary frequency

Yes (n = 38) 100 (83.2, 100) 22.2 (6.4, 47.6) 1.29 — 108 210 100 (83.2, 100) 50.0 (26.0, 74.0) 2.00 —

No (n = 268) 97.2 (92.1, 99.4) 54.4 (46.3, 62.3) 2.13 0.05 74 21 100 (96.6, 100) 47.5 (39.6, 55.5) 1.90 —

Fever (>38.0°C)1

Yes (n = 96) 97.2 (85.5, 99.9) 60.0 (46.5, 72.4) 2.43 0.05 108 14 100 (90.3, 100) 48.3 (35.2, 61.6) 1.94 —

No (n = 209) 97.8 (92.3, 99.7) 46.6 (37.4, 56.0) 1.83 0.05 171 22 96.7 (90.7, 99.3) 48.3 (39.0, 57.7) 1.87 0.07

UTI-specific abdominalpain2

Yes (n = 58) 100 (89.1, 100) 34.6 (17.2, 55.7) 1.53 — 169 209 96.9 (83.8, 99.9) 53.8 (33.4, 73.4) 2.10 0.06

No (n = 248) 96.9 (91.1, 99.4) 53.9 (45.7, 62.1) 2.10 0.06 69 21 100 (96.2, 100) 47.4 (39.2, 55.6) 1.90 —

Diabetes

Yes (n = 74) 97.1 (84.7, 99.9) 57.5 (40.9, 73) 2.28 0.05 76 17 100 (89.7, 100) 50.0 (33.8, 66.2) 2.00 —

No (n = 232) 97.9 (92.5, 99.7) 49.3 (40.7, 57.9) 1.93 0.04 137 56 97.9 (92.5, 99.7) 55.8 (47.1, 64.2) 2.21 0.04

Renal insufficiency

Yes (n = 96) 95.1 (83.5, 99.4) 45.5 (32.0, 59.4) 1.74 0.11 74 21 100 (91.4, 100) 49.1 (35.4, 62.9) 2.00 —

No (n = 210) 98.9 (93.8, 100) 53.7 (44.4, 62.7) 2.13 0.02 108 54 97.7 (91.9, 99.7) 54.5 (45.2, 63.5) 2.15 0.04

Immunosuppression

Yes (n = 110) 100 (90.5, 100) 57.5 (45.4, 69.0) 2.35 — 74 24 97.3 (85.8, 99.9) 53.4 (41.4, 65.2) 2.09 0.05

No (n = 196) 96.7 (90.7, 99.3) 46.7 (36.9, 56.7) 1.81 0.07 151 21 97.8 (92.3, 99.7) 46.7 (36.9, 56.7) 1.83 0.051One patient in the validation set did not have sufficient data to determine if temperature > 38 0°C. 2This includes lower abdominal, suprapubic, and flank pain.Abbreviations: Bact: UFC bacterium counts; CI: confidence interval; Lc: UFC leucocyte counts; LR: likelihood ratio; UFC: urine flow cytometry; UTI: urinarytract infection.

8 Disease Markers

patients who have a combination of the different attributesfor which criteria were presented in this study. In general,if more than one applicable category with adapted cut-offexists, we recommend using the one with the lowest cut-offfor a safe exclusion of future significant bacterial growth inurine culture. Prospective studies are needed to address theimpact of the diagnostic accuracy for significant bacterialgrowth in urine culture for UFC parameters in the settingin which the clinician greatly suspects a UTI. Moreover,our design only enabled us to include people for whom aurine culture was obtained in ordinary clinical practice. Ina prospective design, a urine culture would have beenobtained for all patients with a UFC. As the patients fromthe examined presentations overlap, the results are not inde-pendent and cannot be combined to calculate (for instance)the positive likelihood ratio to rule out significant bacterialgrowth in urine culture in an immunosuppressed, female,older patient.

5. Conclusions

Clinicians should be aware that in specific patient groups,e.g., in females, the predictive accuracy of UFC leucocyteand UFC bacterium counts to predict significant bacterialgrowth in urine culture is diminished and that the predictiveaccuracy of cut-offs depends on the clinical setting. On thebasis of our findings, UFC can be used to rule out significantbacterial growth in urine culture with high sensitivity by theuse of the cut-offs for UFC bacterium counts ≤ 125/μL andUFC leucocyte counts ≤ 17/μL—independently of the clinicalpresentation, sex, age, or comorbidities such as diabetes,renal insufficiency, and immunosuppression. Even higherUFC leucocyte cut-offs (169/μL, 169/μL, and 205/μL) are jus-tified in females, people aged ≤65 years, and dysuric patients.

Data Availability

The statistical data used to support the findings of this studyare available from the corresponding author upon request.

Conflicts of Interest

WEH received speaker honoraria from the AO FoundationZürich and research funding from Mundipharma MedicalCompany, Basel. All other authors have nothing to disclose.

Acknowledgments

MM was funded by the Bangerter Foundation and the SwissAcademy of Medical Sciences through the “Young Talents inClinical Research” grant (TCR 14/17).

References

[1] S. Ozdem, T. Bayraktar, C. Oktay, R. Sari, and M. Gültekin,“The prevalence of asymptomatic pyuria in diabetic patients:comparison of the Sysmex UF-100 automated urinalysis ana-lyzer with Fuchs–Rosenthal hemacytometer,” Clinical Bio-chemistry, vol. 39, no. 9, pp. 873–878, 2006.

[2] T. A. Rowe and M. Juthani-Mehta, “Urinary tract infection inolder adults,” Aging Health, vol. 9, no. 5, pp. 519–528, 2013.

[3] F. L. Ruben, S. R. Dearwater, C. W. Norden et al., “Clinicalinfections in the noninstitutionalized geriatric age group:methods utilized and incidence of infections. The PittsburghGood Health Study,” American Journal of Epidemiology,vol. 141, no. 2, pp. 145–157, 1995.

[4] B. Foxman, “The epidemiology of urinary tract infection,”Nature Reviews Urology, vol. 7, no. 12, pp. 653–660, 2010.

[5] O. P. Kalra and A. Raizada, “Approach to a patient with uro-sepsis,” Journal of Global Infectious Diseases, vol. 1, no. 1,pp. 57–63, 2009.

[6] L. M. A. J. Muller, K. J. Gorter, E. Hak et al., “Increased risk ofcommon infections in patients with type 1 and type 2 diabetesmellitus,” Clinical Infectious Diseases, vol. 41, no. 3, pp. 281–288, 2005.

[7] B. R. Shah and J. E. Hux, “Quantifying the risk of infectiousdiseases for people with diabetes,” Diabetes Care, vol. 26,no. 2, pp. 510–513, 2003.

[8] G. Schmiemann, E. Kniehl, K. Gebhardt, M.M.Matejczyk, andE. Hummers-Pradier, “The diagnosis of urinary tract infec-tion,” Deutsches Ärzteblatt International, vol. 107, pp. 361–367, 2010.

[9] J. Delanghe, “New screening diagnostic techniques in urinaly-sis,” Acta Clinica Belgica, vol. 62, no. 3, pp. 155–161, 2014.

[10] J. C. dos Santos, L. P. Weber, and L. R. R. Perez, “Evaluation ofurinalysis parameters to predict urinary-tract infection,” Bra-zilian Journal of Infectious Diseases, vol. 11, no. 5, pp. 479–481, 2007.

[11] S. Jolkkonen, E.-L. Paattiniemi, P. Kärpänoja, andH. Sarkkinen, “Screening of urine samples by flow cytometryreduces the need for culture,” Journal of Clinical Microbiology,vol. 48, no. 9, pp. 3117–3121, 2010.

[12] F. Manoni, L. Fornasiero, M. Ercolin et al., “Cutoff values forbacteria and leukocytes for urine flow cytometer SysmexUF-1000i in urinary tract infections,” Diagnostic Microbiologyand Infectious Disease, vol. 65, no. 2, pp. 103–107, 2009.

[13] Y.-J. Shang, Q.-Q. Wang, J.-R. Zhang et al., “Systematic reviewand meta-analysis of flow cytometry in urinary tract infectionscreening,” Clinica Chimica Acta, vol. 424, pp. 90–95, 2013.

[14] M. Müller, R. Seidenberg, S. K. Schuh et al., “The developmentand validation of different decision-making tools to predicturine culture growth out of urine flow cytometry parameter,”PLoS One, vol. 13, no. 2, article e0193255, 2018.

[15] F. J. de Boer, E. Gieteling, H. van Egmond-Kreileman et al.,“Accurate and fast urinalysis in febrile patients by flow cytom-etry,” Infectious Diseases, vol. 49, no. 5, pp. 380–387, 2017.

[16] G. Martín-Gutiérrez, A. Porras-González, C. Martín-Pérez,J. A. Lepe, and J. Aznar, “Evaluation and optimization of theSysmex UF1000i system for the screening of urinary tractinfection in primary health care elderly patients,” Enferme-dades Infecciosas y Microbiología Clínica, vol. 33, no. 5,pp. 320–323, 2015.

[17] A. Stefanovic, D. Roscoe, R. Ranasinghe et al., “Performanceassessment of urine flow cytometry (UFC) to screen urines toreflex to culture in immunocompetent and immunosup-pressed hosts,” Journal of Medical Microbiology, vol. 66,no. 9, pp. 1308–1315, 2017.

[18] T. Monsen and P. Ryden, “A new concept and a comprehen-sive evaluation of SYSMEX UF-1000i flow cytometer to iden-tify culture-negative urine specimens in patients with UTI,”

9Disease Markers

European Journal of Clinical Microbiology & Infectious Dis-eases, vol. 36, no. 9, pp. 1691–1703, 2017.

[19] S. McGee, “Simplifying likelihood ratios,” Journal of GeneralInternal Medicine, vol. 17, no. 8, pp. 647–650, 2002.

[20] K. Hajian-Tilaki, “The choice of methods in determining theoptimal cut-off value for quantitative diagnostic test evalua-tion,” Statistical Methods in Medical Research, vol. 27, no. 8,pp. 2374–2383, 2017.

[21] P. Kent and M. J. Hancock, “Interpretation of dichotomousoutcomes: sensitivity, specificity, likelihood ratios, andpre-test and post-test probability,” Journal of Physiotherapy,vol. 62, no. 4, pp. 231–233, 2016.

[22] M. Farm, A. J. Siddiqui, L. Onelöv et al., “Age-adjustedD-dimer cut-off leads to more efficient diagnosis of venousthromboembolism in the emergency department: a compari-son of four assays,” Journal of Thrombosis and Haemostasis,vol. 16, no. 5, pp. 866–875, 2018.

[23] L. G. M. Giesen, G. Cousins, B. D. Dimitrov, F. A. van de Laar,and T. Fahey, “Predicting acute uncomplicated urinary tractinfection in women: a systematic review of the diagnostic accu-racy of symptoms and signs,” BMC Family Practice, vol. 11,no. 1, p. 78, 2010.

[24] K. Huysal, Y. U. Budak, A. Ulusoy Karaca et al., “Diagnosticaccuracy of UriSed automated urine microscopic sedimentanalyzer and dipstick parameters in predicting urine culturetest results,” Biochemia Medica, vol. 23, pp. 211–217, 2013.

[25] E.-L. Paattiniemi, S. Karumaa, A.-M. Viita et al., “Analysis ofthe costs for the laboratory of flow cytometry screening ofurine samples before culture,” Infectious Diseases, vol. 49,no. 3, pp. 217–222, 2016.

[26] F. Habibzadeh, P. Habibzadeh, and M. Yadollahie, “On deter-mining the most appropriate test cut-off value: the case of testswith continuous results,” Biochemia Medica, vol. 26, no. 3,pp. 297–307, 2017.

10 Disease Markers

Stem Cells International

Hindawiwww.hindawi.com Volume 2018

Hindawiwww.hindawi.com Volume 2018

MEDIATORSINFLAMMATION

of

EndocrinologyInternational Journal of

Hindawiwww.hindawi.com Volume 2018

Hindawiwww.hindawi.com Volume 2018

Disease Markers

Hindawiwww.hindawi.com Volume 2018

BioMed Research International

OncologyJournal of

Hindawiwww.hindawi.com Volume 2013

Hindawiwww.hindawi.com Volume 2018

Oxidative Medicine and Cellular Longevity

Hindawiwww.hindawi.com Volume 2018

PPAR Research

Hindawi Publishing Corporation http://www.hindawi.com Volume 2013Hindawiwww.hindawi.com

The Scientific World Journal

Volume 2018

Immunology ResearchHindawiwww.hindawi.com Volume 2018

Journal of

ObesityJournal of

Hindawiwww.hindawi.com Volume 2018

Hindawiwww.hindawi.com Volume 2018

Computational and Mathematical Methods in Medicine

Hindawiwww.hindawi.com Volume 2018

Behavioural Neurology

OphthalmologyJournal of

Hindawiwww.hindawi.com Volume 2018

Diabetes ResearchJournal of

Hindawiwww.hindawi.com Volume 2018

Hindawiwww.hindawi.com Volume 2018

Research and TreatmentAIDS

Hindawiwww.hindawi.com Volume 2018

Gastroenterology Research and Practice

Hindawiwww.hindawi.com Volume 2018

Parkinson’s Disease

Evidence-Based Complementary andAlternative Medicine

Volume 2018Hindawiwww.hindawi.com

Submit your manuscripts atwww.hindawi.com